I've tried a lot of dictation tools, and almost all of them ship your audio off to someone's cloud. That always bugged me. So when I found Inaudio, a tiny desktop app that does all of it locally, I had to share it.
You press a keyboard shortcut and talk. It types. And it keeps a searchable history of everything you've dictated, so you can go back and grab that one sentence from yesterday.
It works the other way too. Paste text or copy something to your clipboard, and it reads it aloud with on-device speech synthesis. Great for proofreading your own writing (you hear the clunky sentences way faster than you see them).
Under the hood it's Electron, React and TypeScript, with sherpa-onnx doing the speech work. You download a model once, then everything runs on your CPU. One of the options is NVIDIA's Nemotron streaming ASR model for English, tuned for about 560 ms latency.
Is it perfect? No. This is a brand new project with only a handful of stars, so expect rough edges. And text insertion and permissions behave differently on Windows, macOS and Linux.
But if you want private dictation without a subscription, it's worth a try. And honestly, it's a great time to jump in and contribute while the project is still small.